See why a system is risky
Each score is broken down into eight factors, such as data sensitivity, autonomy and EU AI Act tier, so you can see exactly what drives it.
Assess
Know the risk of every AI system, continuously.
Qastel classifies and scores every AI system it finds, shows what it is used for and by which teams, and lets you review the results and act on them in Qastel Chat.
AI risk changes every week. Your assessment should too.
A risk review done once, when a tool is bought, is out of date within months. Qastel scores every AI system from what it sees in your estate, and again whenever something changes.
Every system classified
A risk score from 0 to 100 on eight dimensions, with the EU AI Act tier assessed automatically.
Every point explained
The score breaks down into the gaps behind it. An unknown counts as at risk, so a blank never lowers the score.
Every gap becomes an action
One governance score for the whole estate, and each open action shows the points it adds.
Risk classification
Every AI system classified, with the reasoning shown.
Each system gets a risk score built from eight weighted dimensions: governance state, autonomy, data sensitivity, EU AI Act tier, policy coverage, data residency, DPA status and personal-account use.
- EU AI Act tier assessed automatically
- Four autonomy levels, from Observe to Act autonomously
- A change made by a person is never overwritten by later automatic assessment
Usage
See what AI is used for, and by which teams.
Qastel groups AI use by business topic, such as HR, Finance, Legal, Sales or software engineering, and flags the sensitive data sent along the way. The People directory shows that use team by team, down to the personal accounts.
- Use grouped by business topic, once an admin turns on prompt analysis
- People filtered by team, with the AI systems each person used
- Sensitive data in prompts flagged by category, with prompts never stored
Qastel Chat
A chat assistant that takes action on your AI governance.
Ask Qastel Chat about your AI estate in plain language: which systems are risky, who uses them, what to fix first. When a change is needed, such as assigning an owner, marking a system as sanctioned or drafting a policy, Chat prepares it and applies it once you approve.
- Answers from your own data, with charts and its sources
- Assigns owners, changes governance states, snoozes actions and drafts policies
- Nothing changes until a person approves, and every change lands in the audit log
The Qastel difference
What you get that you do not get elsewhere.
Investigate and act through Qastel Chat
Ask about your AI usage in plain language, then handle what you find in the same conversation: assign an owner, change a system's status or draft a policy, each applied once you approve.
Flags new risks as they appear
A new system or a new gap becomes an action in your Inbox, showing the points it is worth. A Guardian can review your AI activity and propose the policies that close it, for you to approve.
Frequently asked questions
How is the risk score calculated?
From explicit rules applied to what Qastel knows about each system, across eight weighted dimensions. Nothing is a black box: the score breaks down into every gap that produced it, the evidence behind each one and exactly how much it cost.
What is the difference between the risk score and the governance score?
The risk score belongs to one AI system: lower is better, and it says how risky that system is. The governance score belongs to the whole estate: higher is better, and it says how much of the estate is reviewed, owned and covered by policy. Closing the actions it lists is how you raise it.
Does assessment require input from system owners?
No questionnaire is required to get a first result: assessment starts from what discovery already found. Owners can add context such as purpose, data categories and DPA references, and that context refines the score, but an unanswered system is still assessed rather than left blank.
What can Qastel Chat change?
Chat can propose an owner, a governance state or a classification for an AI system, add a note, dismiss, snooze or reopen an action, and draft a policy. Each proposal is a card you approve or deny, and your permissions are checked again when you approve. A policy draft is always created disabled.
See Assess in action.
Get a 30-minute overview of how Qastel discovers, assesses and controls the AI used across a business.